Core concepts
ContentsVision has a small, consistent vocabulary. Learn these eight words and most of the product and the codebase will make sense. For a fuller alphabetical list, see the glossary.
The entity hierarchy
Everything hangs off a clear parent-child tree. A team owns claims, a claim has rooms, a room collects jobs, a job produces items, and items can carry photos.
The eight words
Team
An organization. Every other record belongs to a team, and the API enforces that a user only ever sees their own team's data. A team is auto-created on first login from the AdjustSquare token, and it carries the team_token used to authenticate calls to Adjust Square.
User
A person, also auto-provisioned on first login. A user belongs to exactly one team and carries an external_id (their AdjustSquare user id), which becomes the "adjuster id" when claims are submitted.
Claim
The top-level unit of work, mirroring an insurance claim. It holds the claim metadata (claim number, date of loss, ZIP, policy limits, insured name, loss type, coverage type) and, once submitted, the Adjust Square claim key. Item numbers are unique and sequential within a claim.
Room
A subdivision of a claim, such as "Master Bedroom" or "Garage". Media is always uploaded into a specific room. A room is just a labeled bucket that collects jobs.
Job (processing job)
One upload and its processing run. A 4-minute video is one job; a batch of 30 photos is one job; a voice memo is one job. A job has a source_type (video, photo, or audio), a status that advances through the pipeline, and live progress_pct. All the items from one upload share a job.
Item (inventory item)
A single line of the inventory: a 65-inch QLED TV, six dining chairs, a leather sofa. Items carry the inventory fields (name, quantity, condition, age, brand, model, serial, damage type, notes), AI metadata (confidence, evidence), review flags (reviewed, flagged, approved), a claim-unique item_number, and links to photos.
Confidence and flags
Every AI-produced item has a confidence score from 0.0 to 1.0. Items below 0.6 are automatically flagged for attention. Review state has three booleans: reviewed (a human looked at it), flagged (needs attention), and approved (cleared for export and submission).
Credits (billing)
Usage is billed to the customer in credits, not dollars. Video and audio are charged per minute (rounded up); photos are charged per AI "request" (one request per group analyzed). This is entirely separate from the internal dollar cost of the AI tokens, which is tracked for observability. Every instance meters usage this way. Some instances also turn on prepaid billing (ENABLE_BILLING): the team buys credits up front through an external billing service, uploads are blocked when the balance is too low, and finished jobs are charged against it. Enterprise / invoiced instances leave it off and only meter. See billing.
Two photo concepts that are easy to confuse
| Term | What it is | Table |
|---|---|---|
| Staged photo | A raw uploaded photo waiting to be grouped before the AI runs. The user merges shots of the same item into groups here. | staged_photos |
| Item photo | A photo attached to a finished inventory item (the AI's crop, the source photo, extra angles). The source of truth for what an item looks like. | item_photos |
A worked example
A specialist visits a house after a kitchen fire. In ContentsVision they:
- Create a claim
CLM-4471, loss type "Fire".- Add two rooms: "Kitchen" and "Living Room".
- Upload a 3-minute walkthrough video to the Kitchen. That becomes one job.
- The pipeline produces 22 items (refrigerator, microwave, 6 dining chairs, ...), each with a confidence score and a cropped photo.
- The specialist reviews, fixes a couple of names, and approves 20 of them.
- They submit the claim. The 20 approved items and their photos are pushed to Adjust Square, and the team is billed 30 credits (a 3-minute video at 10 credits per minute).
Next: see this play out screen by screen in the user journey.